Dango233
Dango233
Same here. Would be great if youcould help update the pretrained model
Great work! @MatthieuTPHR I was able to get a +60% speed up on a A40 on unet But this seems to **break torch.jit.trace**, I'm getting this error: `RuntimeError: unsupported output...
Understood if we want gradient computed. For forward pass only jit, will fixing the output type of the op work? The int output of the op is where jit breaks....
I have exactly the same problem... Any clue?
I used a model trained under the CompVis/Stable-diffusion format. I got it converted to diffusers format using this conversion script: https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_stable_diffusion_to_diffusers.py
Seems that the config saved using diffusers' `save_pretrained` method will have this problem
> https://github.com/mindspore-ai/models/tree/master/research/mm/wukong
+1 for this - not supporting prefix in `/v1/chat/completion` for me is the largest gap between llama.cpp vs common API providers & lmstudio...
Probelm fixed. > > @Dango233 Hello, may I ask what this prompt means: "hot attention enabled: running without shot embedding." it's a debug info - if a ckpt comes with...
Oh hmmm. Try not to use torch compile for now. I'll look at if we can use compile at all.